HHS Public Access Author manuscript Author ManuscriptAuthor Manuscript Author Nat Biotechnol Manuscript Author . Author Manuscript Author manuscript; available in PMC 2018 August 19. Published in final edited form as: Nat Biotechnol. 2018 March ; 36(3): 272–281. doi:10.1038/nbt.4072. Recon3D: A Resource Enabling A Three-Dimensional View of Gene Variation in Human Metabolism Elizabeth Brunka,b, Swagatika Sahooc,d, Daniel C. Zielinskia, Ali Altunkayae, Andreas Drägerf, Nathan Miha, Francesco Gattoa,g, Avlant Nilssong, German Andres Preciat Gonzalezc, Maike Kathrin Aurichc, Andreas Prliće, Anand Sastrya, Anna D. Danielsdottirc, Almut Heinkenc, Alberto Noronhac, Peter W. Rosee, Stephen K. Burleye,h, Ronan M.T. Flemingc, Jens Nielsenb,g, Ines Thiele*,c, and Bernhard O. Palsson*,a,b aDepartment of Bioengineering, University of California San Diego CA 92093 bThe Novo Nordisk Foundation Center for Biosustainability, Technical University of Denmark, 2800 Lyngby, Denmark cLuxembourg Centre for Systems Biomedicine, University of Luxembourg, Campus Belval, Esch- Sur-Alzette, Luxembourg eRCSB Protein Data Bank, San Diego Supercomputer Center, University of California, San Diego, La Jolla, CA 92093, USA fApplied Bioinformatics Group, Center for Bioinformatics Tübingen (ZBIT), University of Tübingen, 72076 Tübingen, Germany Users may view, print, copy, and download text and data-mine the content in such documents, for the purposes of academic research, subject always to the full Conditions of use: http://www.nature.com/authors/editorial_policies/license.html#terms *correspondence should be addressed to: I.T. ([email protected]) and B.O.P ([email protected]). dCurrent address: Department of Chemical Engineering, Indian Institute of Technology Madras, India 600036 AUTHOR CONTRIBUTIONS Conceptualization, E.B., I.T., D.Z.; Methodology, (Reconstruction of metabolic network: S.S., IT, RMTF, ADD, AH, MKA ; Reconstruction of GEM-PRO: EB, NM AS; 3D-hotspot analysis: EB, AP, AS, PWR; Machine learning: D.Z.; PDB visualization: AA, AP, AD, RMTF, SKB; atom-atom mapping: GAPG, RMTF; Model testing and validation: IT, RMTF, SS, MKA, DZ, AN, FG; Cell- specific and infant model simulations: MKA, AN, FG); Investigation, EB., DZ, GAPG; Writing – Original Draft, EB, BOP; Writing – Review & Editing: all authors; Funding Acquisition: IT, RMTF, SKB, JN, and BOP; Resources, IT, RMTF, SKB, JN, and BOP; Supervision: IT, RMTF, SKB, and BOP. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests. DATA AND CODE AVAILABILITY Recon3D is available available as a metabolic reconstruction at http://vmh.life. To facilitate the future use of the Recon 3D GEM-PRO model, the procedure to collect sequence and structure information as described above has been consolidated into a shareable JSON file, which we call the “minimal” GEM-PRO needed to start structural analyses. This model assigns a single representative structure per gene in the reconstructed metabolic model, and is available at https:// github.com/SBRG/Recon3D. The accompanying software package required for reading and working with the GEM-PRO JSON is available at https://github.com/SBRG/ssbio. This entire repository can be cloned to a user’s computer and contains Jupyter notebooks in the root directory to guide a user through the content available in the Recon 3D GEM-PRO model (Recon3D_GP - Loading and Exploring the GEM-PRO.ipynb) as well as to update the model with revised sequence information or newly deposited structures in the PDB (Recon3D_GP - Updating the GEM-PRO.ipynb). This repository also includes all sequence and structure files mapped per gene, metadata downloaded through UniProt and the PDB, as well as the ability to rerun the QC/QA pipeline with different parameters such as sequence identity and resolution cutoffs. These notebooks also include basic visualization features enabled with the NGL viewer package69. All other scripts related to gene deletion simulations and infant growth simulations can be found at https://github.com/SBRG/ Recon3D. Brunk et al. Page 2 gDepartment of Biology and Biological Engineering, Chalmers University of Technology, Sweden Author ManuscriptAuthor Manuscript Author Manuscript Author Manuscript Author hDepartment of Chemistry and Chemical Biology, Center for Integrative Proteomics Research, Institute for Quantitative Biomedicine, and Rutgers Cancer Institute of New Jersey, Rutgers, The State University of New Jersey, Piscataway, NJ 08854, USA Abstract Genome-scale network reconstructions have helped uncover the molecular basis of metabolism. Here we present Recon3D, a computational resource that includes three-dimensional (3D) metabolite and protein structure data and enables integrated analyses of metabolic functions in humans. We use Recon3D to functionally characterize mutations associated with disease, and identify metabolic response signatures that are caused by exposure to certain drugs. Recon3D represents the most comprehensive human metabolic network model to date, accounting for 3,288 open reading frames (representing 17% of functionally annotated human genes), 13,543 metabolic reactions involving 4,140 unique metabolites, and 12,890 protein structures. These data provide a unique resource for investigating molecular mechanisms of human metabolism. Recon3D is available at http://vmh.life. It is widely recognized that progress in the biomedical sciences is hampered by the difficulty of integrating multiple disparate data types to obtain a coherent understanding of physiological and disease states1. A genome-scale network reconstruction represents a curated knowledge-base containing many different data types and sources, including high- quality genome annotation, assessment of biochemical properties of gene products, and a wide array of physiological functional information. Computational genome-scale models integrate large-scale omics data from these knowledge-bases to aid in the interpretation and prediction of biological functions2. In recent years, human metabolic network reconstructions3–6 have generated insights into inborn errors of metabolism7, cancer8 and human microbiome co-metabolism9,10. Using metabolic reconstructions, information about chemical reactions is stored and continually updated in a standardized biochemical and genetic representation through a well- established process11. Over the past ten years, updating the human metabolic network reconstruction has focused on expansion of metabolic reaction coverage. From the first human reconstruction, Recon14, to the most recent version, Recon23, the content has been expanded from 1,496 genes (corresponding to 3,311 reactions) to 1,675 genes (7,785 reactions). Various other reconstructions have been released and community driven-efforts have been made to ensure interoperability of these resources3,5. Our knowledge about human metabolism is continuously increasing and the deluge of ‘omics’ data provides ample opportunity for updating current knowledge-bases of human metabolism3–6. In addition to updating metabolic coverage, expanding human reconstructions to include different types, such as metabolite and protein structures as well as atom- transitions, thereby enables a broader scope of biomedical questions to be addressed. Historically, systems biology has focused on characterizing catalytic or regulatory roles of proteins in metabolism without placing emphasis on the three-dimensional structure of the proteins themselves. For example, studies on genetic variation have mainly focused on Nat Biotechnol. Author manuscript; available in PMC 2018 August 19. Brunk et al. Page 3 frequency of occurrence12 or sequence-based attributes13. Only recently have mutations Author ManuscriptAuthor Manuscript Author Manuscript Author Manuscript Author been explored in the context of their three-dimensional location or spatial relationship14–17. Exploring mutations in 3D extends beyond nucleotide sequence identity18, as mutations that may be far away from each other in linear sequence may actually be proximal in the folded state. In recent years, increasingly accessible protein and metabolite data have enabled the progression of systems biology to a 3D perspective. In one study, protein structures were mapped to the metabolic network of Escherichia coli, to reveal the role of ribosome pausing in co-translational protein folding19. In another study, human population variation was studied by integrating protein structures with the human erythrocyte metabolic network to understand the adverse effects of drugs on genetic variants20 and identify new pathways related to drug perturbation. These studies highlight the value of integrating different types of data to address complex biological questions. We present Recon3D, an updated and expanded human metabolic network reconstruction that integrates pharmacogenomic associations, large-scale phenotypic data, and structural information for both proteins and metabolites. Recon3D contains over 6,000 more reactions than Recon2, all of which were manually curated to remove redundant or blocked reactions. We use Recon3D to prioritize putative disease-causing genetic variants by mapping single nucleotide variants (SNVs) to protein structures. We show that deleterious mutations are more likely to cluster together into functional hotspots than non-deleterious mutations. In contrast to previous models, these mutation hotspots identify ACAT1 as a cancer-related gene. Furthermore, we demonstrate how structural information can be used to investigate the potential